Missing Session-Based Tracing API #398
Description
Title: Missing session-based tracing API for complex multi-step agent workflows
Description:
I'm building a resume analysis agent that processes entire sessions with multiple sections (skills, education, experience, projects) in sequence. The current tracing API only supports individual function tracing, but I need session-level correlation to track the complete workflow and understand how different sections relate to each other.
Problem:
My agent has a complex workflow where it:
- Processes multiple sections in sequence
- Makes decisions at each step (accept/reject/retry)
- Sometimes pauses for human clarification
- Maintains session state throughout the process
Currently, I have to manually track session state and correlate traces:
session = {
"session_id": session_id,
"created_at": datetime.now().isoformat(),
"sections": parsed_sections,
"job_analysis": job_analysis,
"section_analyses": {},
"accepted_changes": {},
"needs_clarification": False,
"pending_clarifications": {},
"current_section_index": 0
}Expected Behavior:
I was hoping for a session-level API that would let me correlate all the spans within a single session, something like:
with judgment.session(session_id="resume_analysis_123") as session:
session.add_metadata("total_sections", len(sections))
session.correlate_spans("section_analysis")
# All spans within this context would be correlated
for section in sections:
with session.span(f"analyze_{section.type}"):
# This span would be part of the session
analyze_section(section)Current Behavior:
Only individual @judgment.observe() decorators are available, requiring manual session state management and correlation. This makes it difficult to:
- Track the complete session lifecycle
- Correlate related spans across different sections
- Understand session-level metrics and performance
- Debug issues that span multiple sections
Impact:
Without session-level tracing, I can't effectively monitor and debug complex agent workflows. This is particularly important for production systems where understanding the complete user journey is crucial.